Research On Algorithms For Evaluating Breast Cancer Stromal Tumor Infiltrating Lymphocytes Based On Region Segmentation And Nuclear Segmentation Classification

The histological evaluation of stromal tumor infiltrating lymphocytes(s TILs)can be used as a surrogate for the host immune response and has been shown to have prognostic and potential chemical predictive effects in HER-2 positive and triple negative breast cancer(TNBC).The current practice of manual evaluation can easily lead to differences within and between observers.In order to meet this challenge,this paper proposes a method based on region segmentation and nuclear segmentation classification to evaluate s TILs of H&E stained breast cancer pathological images,with the following main work:1.By analyzing the correlation between s TILs score and s TILs density,a linear regression equation of s TILs score and s TILs density was constructed.By calculating the s TILs density on the pathological image of breast cancer,the s TILs score can be calculated from the linear relationship on the pathological image of breast cancer;2.Based on the Link Net network model to segment the stromal region of breast cancer pathological images,and compared with the segmentation methods based on UNet and Pix2 Pix.From comparative experiments,the Link Net-based network model can effectively segment the stromal region of breast cancer pathological images,and its DICE index reaches 0.9274;3.Based on the Hover-Net network model for nuclear segmentation and classification of breast cancer pathology images,a machine learning-based nuclear classification algorithm was proposed for its problems.From the comparative experiment,Random Forest classifier better than other nuclear classification methods,its accuracy index reached 0.837,and the best F1-score index was obtained when classifying epithelial cells,lymphocytes,and stromal cells,which were 0.744,0.752,and 0.639,respectively.4.Calculate s TILs scores for breast cancer pathological images based on region segmentation and nuclear segmentation classification methods,and analyzes the correlation between the pathologist’s score and the calculated s TILs score,among them,the pathologist’s consensus score and the calculated s TILs score are highly correlated,and the corresponding Spearman correlation coefficient r=0.869(p<0.001)was higher than the Spearman correlation coefficient r=0.851(p<0.001)of the scores among physiologists,which verifies the validity of the proposed method based on region segmentation and nuclear segmentation classification for calculated s TILs score.Finally,this paper summarizes the above work and suggests directions for improvement based on its practical results in the evaluation of s TILs on breast cancer pathology images.